Forecasting Method Of Nonlinear Time Series Of Slope Deformat Using Gaussian Process

With the continually development of national economy and the national attention of water conservancy project, water conservancy project construction is in the ascendant.Water conservancy project construction will always meet high slope stability problem, the problem of landslide induced by the reservoir impoundment is getting more and more attention. Slope safety seriously influences people’s life and property security, Slope deformation time series prediction is an important research field in water conservancy engineering discipline. The slope systemis a complex nonlinear dynamic system,its stability is under the comprehensive influence of various factors such as hydrogeological conditions, surrounding environment and human engineering activities. Its failure mechanism is extraordinarily complex, which making the slope deformation time series forecast and stability evaluation accurately encountering great difficulties. Thus, studying nonlinear time series of slope deformation has important realistic meaning and obvious research value.Gussian process (GP) is a newly developed machine learning method based on the strict statistical learning theory. GP has excellent capability for solving the highly nonlinear problems with small samples and high dimension. In this paper, GP will be introduced to the prediction of slope displacement, the main research work and the results as follows:(1) As we all know, the prediction accuracy of traditional time series prediction method based on linear model is not high, the neural network appeared to over-learning and the optimal network structure with its parameters are difficult to be determined, kernel functions and the parameters of support vector machine (SVM) are also difficult to be determine reasonably.So we put forward a nonlinear time series forecasting model based on gaussian process regression(GPR) model.The classic example research results show that the method is feasible. Compared with the traditional methods, its extrapolated predictive precision has certain advantages. (2)We proposed the slope deformation nonlinear time series forecasting model based on gaussian process regression(GPR) model.Through studying the permanent ship lock high slope of Three Gorges Project,landslide of Wolongsi and high slope of Longtan hydropower station inlet study,the research results show that the forecast precision of this method is high, parameters can be obtained adaptively. From the research results we can also find that this method has good applicability for slope nonlinear time series prediction(3) Owing to the limitation of past prediction research of slope deformation time series outside the science evaluation means forecasting results, extrapolating forecasting results become difficult to evaluate scientificly and the reasonable forecasting step length become hard to define. By the forecasting the variance of GPR model,we put forward the the relative uncertainty coefficient (RUC) of slope deformation time series forecasting results. Through describing quantitatively the uncertainty of the forecasting results, the uncertain management level standard of slope deformation time series forecasting results was established, which provided scientific basis for chosing reasonable slope deformation time series forecasting step length.